Harnessing the potential of IoT for quality control in manufacturing

Posted on

This case study will address:

  • Exploitation of legacy data, software, hardware, facilities, skills or knowledge;
  • Bequeathing a rich legacy of new or improved data sources, technology, infrastructure, facilities, practices, skills or partnerships.
  • Development of skills, including: technical, managerial and methods

The aim of this Pitch-In-supported project was to investigate the potential of IoT to significantly improve manufacturing process quality control. 

Manufacturing companies typically perform quality prediction control using traditional statistical process control – however this can be a labour-intensive, manual process, which involves producing a limited size of samples for each batch and conducting manual quality checks. 

With automated IoT data streams, quality engineers brought together through this Pitch-In project, were able to observe finer-grained, sensor-based process data such as temperature, pressure and vibration, then relate these to quality outcomes automatically. The result produced significantly more efficient ways to check quality.

Led by the Manufacturing Analytics Research Group in University of Cambridge, this Pitch-In project featured a collaboration with several manufacturers including Tinsley Bridge, Federal Mogul and Scania as well as research institutes at the University of Sheffield and the Advanced Forming Research Centre (AFRC) at the University of Strathclyde. 

The team came together to find a way to help manufacturers evaluate the potential for IoT data to enhance standard quality optimisation processes through a number of use cases, demonstrating how it can be applied. 

What were the problems or barriers?

There is often a lack of understanding among manufacturing companies of the potential for IoT solutions to add value to their processes.

In particular, factories are often missing out on the opportunity for IoT to make significant efficiencies, for example, they do not know how to use IoT in building predictions for quality checking. They are rarely aware of the benefits that IoT can bring or how they can go about achieving those benefits.

What did you do?

Predicting quality in manufacturing

This Pitch-In project enabled researchers to come together to use state-of-the-art IoT data capture and AI algorithms to predict quality in manufacturing. In this way, they were able to detect anomalies in production without the need to measure quality manually. 

The first project goal was to find a way of making IoT-based data work alongside traditional, batch-level quality control data that had been obtained and recorded manually during manufacturing processes. A new way was needed to use historical knowledge and data and feed it into the new, IoT-based machine learning approach, where checks are item-level, and based on sensor data generated during the process.

Machine learning algorithms were developed that used data from a number of manufacturing environments to provide increased prediction accuracy in quality checks and self-learning capability. 

Another improvement was the provision of uncertainty measures and confidence levels that underpin predictions, to increase manufacturers’ trust in machine learning. 

Finally, researchers developed a machine-learning testbed that reflects the data flow of a multi-stage manufacturing process. The testbed included an experimentation software package, which allowed different use case applications to be automatically fine-tuned. 

What was the result? 

Pitch-In support enabled a number of positive outcomes relating to quality checking in manufacturing processes:

  • Having developed a new method for process condition monitoring, this was tested at a legacy electrophoresis painting plant at Tinsley Bridge where an IoT system had been recently deployed. The company has now adopted the method in its plant.   
  • The dataset was made available on Kaggle, a website used by researchers to test algorithms. Normally researchers are forced to use information that is not directly relevant to manufacturing when they are trying to develop AI for their own applications. By making these new, industry-led data sets available, this will hopefully attract researchers into manufacturing. In the first three months, the data sets were downloaded nearly 500 times, including many from India and the USA. View the project’s dedicated page on the Kaggle website for a live update.
  • The methods and findings from this Pitch-In-supported project were submitted for publication to two academic journals. 
  • The algorithms have been made publicly available through GitHub, an online space for sharing open-source data, so that any researcher can use them and apply them in any other manufacturing organisation.
  • A video was created to show how the algorithm works and demonstrate its potential to manufacturing organisations wanting to adapt it.

Lessons learnt

  • Knowledge exchange: Knowledge exchange between multiple academic partners and industry partners was highly successful (SCANIA, Tinsley Bridge, Federal Mogul and the Universities of Sheffield and Cambridge). Industry partners introduced their real-world case studies and explored the IoT barriers with academic partners. The IoT techniques developed by academic partners were demonstrated to a mixed community including research groups and industry users.
  • Connecting capabilities: Connecting capabilities was crucial to the success of this Pitch-In project. The University of Sheffield provided its hardware and software expertise in developing the IoT data capture; the University of Cambridge provided advanced data analytics capabilities; Tinsley Bridge provided the factory test environment. By combining capabilities and expertise, parties were able to overcome challenges together and work towards improved quality processes.
  • Effective simplicity builds trust: AI and machine learning algorithms need to be easy to interpret and as minimalist as possible, so that manufacturers can understand them and feel more confident in adapting them.
  • Public data sets are a significant enabler: There are challenges for manufacturers in analysing real manufacturing data as there aren’t many publicly available data sets. In this case, Tinsley Bridge permitted the publishing of their data sets, which enables future research.
  • On-site contact is beneficial: Physical visits to the factory were limited by the COVID-lock down, so it would have been beneficial to have more face-to-face visits for academic and industry partners. In particular, visiting the manufacturing shop floor would help researchers understand the problems. Likewise, visiting the lab experiment would help industry partners to understand the analytical workflow.

What’s next?

Further research and industrial projects are planned to develop the quality control algorithms and work on more use cases. 

More journal publications are planned to share the data sets with the research community. The project and its outcomes will also feature in IoT and machine learning conferences, when relevant events can be scheduled in the academic calendar.

Dr Alexandra Brintrup, University of Cambridge said:

“Thanks to the support from Pitch-In, this project in quality analytics demonstrated how IoT data can be used to improve quality control processes. It also enabled us to devise novel AI methods to integrate existing statistical process control data and know-how with new, automated approaches.”

Contact: If you are a manufacturing organisation and are  interested in collaborating on a project related to using IoT for quality control purposes, get in touch with Dr Alexandra Brintrup ab702@cam.ac.uk.

< Share >

Your cookie choices - Our website uses cookies to help give you a better experience. By accepting all cookies, you agree to the storing of cookies on your device.